I use a monthly panel of provincially-collected central government revenues and conflict fatalities to estimate government revenues lost due to conflict in Afghanistan since 2005. I identify causal effects by instrumenting for conflict using pre-sample ethno-linguistic share. Headline estimates are very large, implying total revenue losses since 2005 of $3bn, and future revenue gains from peace of about 6 percent of GDP per year. Reduced collection efficiency, rather than lower economic activity, appears to be the key channel. OLS estimates understate the causal effect by a factor of four. Comparing to estimates from Powell's (2017) generalized synthetic control method suggests that this bias results from omitted variables and measurement error in equal share. The findings underscore the considerable economic loss due to conflict, and the importance of careful identification in measuring this loss.